dranger003's picture
Update README.md
48c38b0 verified
metadata
license: llama2
library_name: gguf
pipeline_tag: text-generation

GGUF importance matrix (imatrix) quants for https://huggingface.co/codellama/CodeLlama-70b-Instruct-hf
The importance matrix was trained for 100K tokens (200 batches of 512 tokens) using wiki.train.raw.

NOTE: The template for this model is very sensitive and must be set very precisely.
All whitespace is intended, and special tokens <s> and <step> must be encoded properly, i.e. 1 and 32015 respectively.

Layers Context Template
80
4096
<s>Source: system

{instructions} <step> Source: user

{prompt} <step> Source: assistant
Destination: user

{response}